giuseros commented on a change in pull request #7785:
URL: https://github.com/apache/tvm/pull/7785#discussion_r614807239



##########
File path: src/target/source/codegen_c_host.cc
##########
@@ -40,13 +40,16 @@ namespace codegen {
 
 CodeGenCHost::CodeGenCHost() { module_name_ = 
GetUniqueName("__tvm_module_ctx"); }
 
-void CodeGenCHost::Init(bool output_ssa, bool emit_asserts, std::string 
target_str) {
+void CodeGenCHost::Init(bool output_ssa, bool emit_asserts, bool 
is_aot_executor,
+                        std::string target_str) {

Review comment:
       Hi @areusch ,
   As I said on the discuss post, I agree about having a separate discussion on 
Memory management, PackedFunc and firmware-facing API. I would say that none of 
this is covered in this PR. In this PR, we have:
   * Relay->TIR codegen (this is the main part)
   * Stack allocator (this is only an add-on, which can be exposed through the 
usual PlatformAllocator, as for the normal paged allocator)
   * Got rid of function registry when calling the functions in AOT
   * added a minimalistic tvm_runtime_run function in aot_executor.h that calls 
the PackedFunc contained in the tvm_model_t struct. 
   The only way I affect the CRT "interface" in this PR is by adding a new 
implementation of `c_backend_api.h` in `aot_backend_api.c` to get rid of 
`TVMFuncRegisterGlobal` which we don't use anymore. If you wish, I can use 
`crt_backend_api.c` and add a fake implementation of `TVMFuncRegisterGlobal` to 
work around the undefined symbol
   So I would propose to finish this review, and start tackling the others 
problems in different RFCs




-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
[email protected]


Reply via email to